Recently, a renewed interest in thin-film magnetic random access memories (MRAM) has been sparked by the potential application of MRAM to both nonvolatile and volatile memories. FIG. 1 depicts a portion of a conventional MRAM 1. The conventional MRAM includes conventional orthogonal conductor lines 10 and 12, conventional magnetic storage cell 11 and conventional transistor 13. The conventional MRAM 1 utilizes a conventional magnetic tunneling junction (MTJ) stack 11 as a memory cell. Use of a conventional MTJ stack 11 makes it possible to design an MRAM cell with high integration density, high speed, low read power, and soft error rate (SER) immunity. The conductive lines 10 and 12 are used for writing data into the magnetic storage device 11. The MTJ stack 11 is located on the intersection of and between 10 and 12. Conventional conductive line 10 and line 12 are referred to as the conventional word line 10 and the conventional bit line 12, respectively. The names, however, are interchangeable. Other names, such as row line, column line, digit line, and data line, may also be used.
The conventional MTJ 11 stack primarily includes the free layer 1104 with the changeable magnetic vector (not explicitly shown), the pinned layer 1102 with the fixed magnetic vector (not explicitly shown), and the insulator 1103 in between the two magnetic layers 1104 and 1102. The insulator 1103 typically has a thickness that is low enough to allow tunneling of charge carriers between the magnetic layers 1102 and 1104. Layer 1101 is usually a composite of seed layers and an anti-ferromagnetic layer that is strongly coupled to the pinned magnetic layer.
Data is stored in the conventional MTJ stack 11 by applying a magnetic field to the conventional MTJ stack 11. The applied magnetic field has a direction chosen to move the changeable magnetic vector of the free layer 1104 to a selected orientation. During writing, the electrical current I1 flowing in the conventional bit line 12 and I2 flowing in the conventional word line 10 yield two magnetic fields on the free layer 1104. In response to the magnetic fields generated by the currents I1 and I2, the magnetic vector in free layer 1104 is oriented in a particular, stable direction. This direction depends on the direction and amplitude of I1 and I2 and the properties and shape of the free layer 1104. Generally, writing a zero (0) requires the direction of either I1 or I2 to be different than when writing a one (1). Typically, the aligned orientation can be designated a logic 1 or 0, while the misaligned orientation is the opposite, i.e., a logic 0 or 1, respectively.
Stored data is read or sensed by passing a current through the conventional MTJ cell from one magnetic layer to the other. During reading, the conventional transistor 13 is turned on and a small tunneling current flows through the conventional MTJ cell. The amount of the current flowing through the conventional MTJ cell 11 or the voltage drop across the conventional MTJ cell 11 is measured to determine the state of the memory cell. In some designs, the conventional transistor 13 is replaced by a diode, or completely omitted, with the conventional MTJ cell 11 in direct contact with the conventional word line 10.
Although the above conventional MTJ cell 11 can be written using the conventional word line 10 and conventional bit line 12, one of ordinary skill in the art will readily recognize that the amplitude of I1 or I2 is in the order of several milli-Amperes for most designs. Therefore, one of ordinary skill in the art will also recognize that a smaller writing current is desired for many memory applications.
FIG. 2 depicts a portion of a conventional magnetic memory 1′ that has a lower writing current. Similar systems are described in U.S. Pat. Nos. 5,659,499, 5,940,319, 6,211,090, 6,153,443, and U.S. Patent Application Publication No. 2002/0127743. The conventional systems and conventional methods for fabricating the conventional systems disclosed in these references encapsulate bit lines and word lines with soft magnetic cladding layer on the three surfaces not facing MTJ cell 11′. Many of the portions of the conventional memory depicted in FIG. 2 are analogous to those depicted in FIG. 1 and are thus labeled similarly. The system depicted in FIG. 2 includes the conventional MTJ cell 11′, conventional word line 10′ and bit line 12′. The conventional word line 10′ is composed of two parts: a copper core 1001 and a soft magnetic cladding layer 1002. Similarly, the conventional bit line 12′ is composed of two parts: a copper core 1201 and a soft magnetic cladding layer 1202.
Relative to the design in FIG. 1, the soft magnetic cladding layers 1002 and 1202 can concentrate the magnetic flux associated with I1 and I2 onto the MTJ cell 11′ and reduce the magnetic field on the surfaces which are not facing the MTJ cell 11′. Thus, the sot magnetic cladding layers 1002 and 1202 concentrate the flux on the MTJ that makes up the MTJ cell 11′, making the free layer 1104 easier to program.
Although this approach works well theoretically, one of ordinary skill in the art will readily recognize that the magnetic properties of the portions of the soft cladding layers 1002 and 1202 on the vertical sidewalls of the conventional lines 10′ and 12′, respectively, are hard to control. One of ordinary skill in the art will also recognize that the process of making the conventional word line 10′ and the conventional bit line 12′ is complicated. Formation of the conventional word line 10′ and conventional bit line 12′ including the cladding layers 1002 and 1202, respectively, requires approximately nine thin film deposition steps, five photolithography steps, six etching steps, and one chemical mechanical polishing (CMP) step. Furthermore, none of the processes can be shared with other CMOS processes. Some of the processes, such as the CMP process and a few thin-film deposition and etching processes, need to be tightly controlled in order to achieve the designed performance. Because the wafer surface on which the devices are fabricated is not flat and the portion to be removed is deep in the trenches, the write lines 10′ and 12′ need to be laid out fairly sparsely to accommodate the photolithography process. As a consequence, the density and capacity of memory devices on a chip will be sacrificed if soft magnetic cladding layer 1202 and 1002 is used for the lines 10′ and 12′. This complicated fabrication methods pose significant challenge to scaling to higher densities. Accordingly it is highly desirable to provide an MRAM architecture which is scalable, easy to fabricate, and offers high writing efficiency.
Other aspects of the conventional write lines 10, 10′, 12, and 12′ of the conventional designs depicted in both FIG. 1 and FIG. 2 limit scalability. In these conventional designs, the conventional write lines 10, 10′, 12, and 12′ are mostly made of either aluminum or copper. The current density limits for aluminum and copper are in the order of 1×106 A/cm2 or less. As the line width decreases to increase the memory density, the electromigration current density limit poses severe challenges for scaling.
Other conventional systems attempt to propose different solutions, each of which has its drawbacks. As an example, U.S. Patent Application Publication No. 2002/0080643 proposed that, after a write operation, a reverse current is applied to the write lines to prevent electromigration. But such conventional methods compromise performance by reducing the speed of the memory and add complexities. Thus, it is also highly desirable to have write line made of materials with high reliability in electromigration, which will allows for easy scalability to high density memory arrays.
Conventional thin bit lines, which might be used for smaller or more efficient memories have shortcomings. Thinner conventional bit lines have higher resistances. This adversely affects the performance of the overall memory array. However, there are many conventional methods of overcoming this issue. One common practice is to break up the long bit lines in the memory array into global bit lines that are made of thick metals, and connect the global bit lines into local bit lines that are made of thinner metals, and thus have a higher resistance. Examples of such design are taught by U.S. Pat. No. 6,335,890 and U.S. Patent Application Publication No. 2002/0034117. However, the other problems described above, such as the electromigration are still not overcome.
Similarly, other conventional systems break write lines into segments, each of which is coupled, through a selection transistor, to a global write line having a much lower resistance. For example, U.S. Pat. No. 6,335,890 and U.S. Patent Application Publication No. 2002/0176272 describe systems which break conventional write lines into segments. During writing, only one segment of the conventional write line conducts current.
FIG. 3 depicts such a conventional design 20 incorporating a conventional segmented write line. The conventional design 20 includes conventional MTJs 31 serving as MRAM cells, a conventional segmented magnetic write line 32, conventional global write and return lines 28 and 30, respectively, conventional selection transistors 26, conventional digit lines 24, and conventional bit selection transistors 22. The conventional segmented write line 32 is connected to the conventional global write line 28 and to the conventional global write return line 30, through the conventional section selection transistor 26. Other segments (not shown) would be connected to the conventional global write line 28 and conventional global write return line 30 in a similar manner. For illustration, the section of the conventional segmented write line 32 is connected to four conventional MTJs 31. In a conventional configuration, each conventional MTJ 31 is then connected to a ground line through a conventional bit selection transistor 22. The conventional digit line 24 runs orthogonal to the conventional write line 32 at each conventional MTJ 31.
During writing, the conventional transistor 26 is turned on to allow a write current to flow from the conventional global write line 28, through the segment of the conventional segmented write line 32, to the conventional to global write return line 30. The magnetic field generated by the write current in the conventional segmented write line 32 simultaneously disturbs the magnetic vectors of the free layers of the four conventional MTJs 31 contained in the section. Depending on the magnitude and polarities of currents flowing in the conventional digit lines 24, digital data are written into the conventional MTJs 31. In this way four MRAM cells can be written at the same time.
Although four MRAM cells can be written in parallel, one of ordinary skill in the art will readily recognize that during reading only one of the four bit selection transistors 22 can be turned on. When one of the four bit selection transistors 22 is turned on, the voltage on the conventional global write line 28 (as connected to the conventional segmented write line 32) can be measured to determine the logical state of the selected MTJ 31. Thus, although data in the four cells of the section can be written in parallel, the data stored in the four cells are read in series. One of ordinary skill in the art will readily recognize, therefore, that reading of data stored in the MTJs 31 may be less efficient than desired. Furthermore, other problems described above, such as the electromigration are still not overcome.
Accordingly, what is needed is a system and method for providing a scalable, efficient, low current magnetic memory that improves ease of manufacturing and reliability against electromigration. It would also be desirable to provide an architecture which can support a simpler driver design, and can be read more efficiently. Moreover, it is desirable to arrange the global write lines such that there is minimal magnetic interference induced by the currents flowing in the global write lines on MRAM cells. It would also be desirable to achieve a system and method for providing a high-density nonvolatile MRAM capable of having greater margin in writing as well as efficient reading operations. The present invention addresses such a need.